linux/kernel/bpf/hashtab.c

1309 lines
32 KiB
C
Raw Normal View History

/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
* Copyright (c) 2016 Facebook
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of version 2 of the GNU General Public
* License as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*/
#include <linux/bpf.h>
#include <linux/jhash.h>
#include <linux/filter.h>
#include <linux/rculist_nulls.h>
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
#include "percpu_freelist.h"
#include "bpf_lru_list.h"
#include "map_in_map.h"
struct bucket {
struct hlist_nulls_head head;
raw_spinlock_t lock;
};
struct bpf_htab {
struct bpf_map map;
struct bucket *buckets;
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
void *elems;
union {
struct pcpu_freelist freelist;
struct bpf_lru lru;
};
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
struct htab_elem *__percpu *extra_elems;
atomic_t count; /* number of elements in this hashtable */
u32 n_buckets; /* number of hash buckets */
u32 elem_size; /* size of each element in bytes */
};
/* each htab element is struct htab_elem + key + value */
struct htab_elem {
union {
struct hlist_nulls_node hash_node;
struct {
void *padding;
union {
struct bpf_htab *htab;
struct pcpu_freelist_node fnode;
};
};
};
union {
struct rcu_head rcu;
struct bpf_lru_node lru_node;
};
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
u32 hash;
char key[0] __aligned(8);
};
static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node);
static bool htab_is_lru(const struct bpf_htab *htab)
{
return htab->map.map_type == BPF_MAP_TYPE_LRU_HASH ||
htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
}
static bool htab_is_percpu(const struct bpf_htab *htab)
{
return htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH ||
htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
}
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
static bool htab_is_prealloc(const struct bpf_htab *htab)
{
return !(htab->map.map_flags & BPF_F_NO_PREALLOC);
}
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
static inline void htab_elem_set_ptr(struct htab_elem *l, u32 key_size,
void __percpu *pptr)
{
*(void __percpu **)(l->key + key_size) = pptr;
}
static inline void __percpu *htab_elem_get_ptr(struct htab_elem *l, u32 key_size)
{
return *(void __percpu **)(l->key + key_size);
}
static void *fd_htab_map_get_ptr(const struct bpf_map *map, struct htab_elem *l)
{
return *(void **)(l->key + roundup(map->key_size, 8));
}
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
static struct htab_elem *get_htab_elem(struct bpf_htab *htab, int i)
{
return (struct htab_elem *) (htab->elems + i * htab->elem_size);
}
static void htab_free_elems(struct bpf_htab *htab)
{
int i;
if (!htab_is_percpu(htab))
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
goto free_elems;
for (i = 0; i < htab->map.max_entries; i++) {
void __percpu *pptr;
pptr = htab_elem_get_ptr(get_htab_elem(htab, i),
htab->map.key_size);
free_percpu(pptr);
}
free_elems:
bpf_map_area_free(htab->elems);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
}
static struct htab_elem *prealloc_lru_pop(struct bpf_htab *htab, void *key,
u32 hash)
{
struct bpf_lru_node *node = bpf_lru_pop_free(&htab->lru, hash);
struct htab_elem *l;
if (node) {
l = container_of(node, struct htab_elem, lru_node);
memcpy(l->key, key, htab->map.key_size);
return l;
}
return NULL;
}
static int prealloc_init(struct bpf_htab *htab)
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
{
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
u32 num_entries = htab->map.max_entries;
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
int err = -ENOMEM, i;
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
if (!htab_is_percpu(htab) && !htab_is_lru(htab))
num_entries += num_possible_cpus();
htab->elems = bpf_map_area_alloc(htab->elem_size * num_entries);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
if (!htab->elems)
return -ENOMEM;
if (!htab_is_percpu(htab))
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
goto skip_percpu_elems;
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
for (i = 0; i < num_entries; i++) {
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
u32 size = round_up(htab->map.value_size, 8);
void __percpu *pptr;
pptr = __alloc_percpu_gfp(size, 8, GFP_USER | __GFP_NOWARN);
if (!pptr)
goto free_elems;
htab_elem_set_ptr(get_htab_elem(htab, i), htab->map.key_size,
pptr);
}
skip_percpu_elems:
if (htab_is_lru(htab))
err = bpf_lru_init(&htab->lru,
htab->map.map_flags & BPF_F_NO_COMMON_LRU,
offsetof(struct htab_elem, hash) -
offsetof(struct htab_elem, lru_node),
htab_lru_map_delete_node,
htab);
else
err = pcpu_freelist_init(&htab->freelist);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
if (err)
goto free_elems;
if (htab_is_lru(htab))
bpf_lru_populate(&htab->lru, htab->elems,
offsetof(struct htab_elem, lru_node),
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
htab->elem_size, num_entries);
else
pcpu_freelist_populate(&htab->freelist,
htab->elems + offsetof(struct htab_elem, fnode),
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
htab->elem_size, num_entries);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
return 0;
free_elems:
htab_free_elems(htab);
return err;
}
static void prealloc_destroy(struct bpf_htab *htab)
{
htab_free_elems(htab);
if (htab_is_lru(htab))
bpf_lru_destroy(&htab->lru);
else
pcpu_freelist_destroy(&htab->freelist);
}
static int alloc_extra_elems(struct bpf_htab *htab)
{
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
struct htab_elem *__percpu *pptr, *l_new;
struct pcpu_freelist_node *l;
int cpu;
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
pptr = __alloc_percpu_gfp(sizeof(struct htab_elem *), 8,
GFP_USER | __GFP_NOWARN);
if (!pptr)
return -ENOMEM;
for_each_possible_cpu(cpu) {
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
l = pcpu_freelist_pop(&htab->freelist);
/* pop will succeed, since prealloc_init()
* preallocated extra num_possible_cpus elements
*/
l_new = container_of(l, struct htab_elem, fnode);
*per_cpu_ptr(pptr, cpu) = l_new;
}
htab->extra_elems = pptr;
return 0;
}
/* Called from syscall */
static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
{
bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
/* percpu_lru means each cpu has its own LRU list.
* it is different from BPF_MAP_TYPE_PERCPU_HASH where
* the map's value itself is percpu. percpu_lru has
* nothing to do with the map's value.
*/
bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
struct bpf_htab *htab;
int err, i;
u64 cost;
BUILD_BUG_ON(offsetof(struct htab_elem, htab) !=
offsetof(struct htab_elem, hash_node.pprev));
BUILD_BUG_ON(offsetof(struct htab_elem, fnode.next) !=
offsetof(struct htab_elem, hash_node.pprev));
if (lru && !capable(CAP_SYS_ADMIN))
/* LRU implementation is much complicated than other
* maps. Hence, limit to CAP_SYS_ADMIN for now.
*/
return ERR_PTR(-EPERM);
if (attr->map_flags & ~(BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU))
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
/* reserved bits should not be used */
return ERR_PTR(-EINVAL);
if (!lru && percpu_lru)
return ERR_PTR(-EINVAL);
if (lru && !prealloc)
return ERR_PTR(-ENOTSUPP);
htab = kzalloc(sizeof(*htab), GFP_USER);
if (!htab)
return ERR_PTR(-ENOMEM);
/* mandatory map attributes */
htab->map.map_type = attr->map_type;
htab->map.key_size = attr->key_size;
htab->map.value_size = attr->value_size;
htab->map.max_entries = attr->max_entries;
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
htab->map.map_flags = attr->map_flags;
/* check sanity of attributes.
* value_size == 0 may be allowed in the future to use map as a set
*/
err = -EINVAL;
if (htab->map.max_entries == 0 || htab->map.key_size == 0 ||
htab->map.value_size == 0)
goto free_htab;
if (percpu_lru) {
/* ensure each CPU's lru list has >=1 elements.
* since we are at it, make each lru list has the same
* number of elements.
*/
htab->map.max_entries = roundup(attr->max_entries,
num_possible_cpus());
if (htab->map.max_entries < attr->max_entries)
htab->map.max_entries = rounddown(attr->max_entries,
num_possible_cpus());
}
/* hash table size must be power of 2 */
htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
err = -E2BIG;
if (htab->map.key_size > MAX_BPF_STACK)
/* eBPF programs initialize keys on stack, so they cannot be
* larger than max stack size
*/
goto free_htab;
if (htab->map.value_size >= KMALLOC_MAX_SIZE -
bpf: fix allocation warnings in bpf maps and integer overflow For large map->value_size the user space can trigger memory allocation warnings like: WARNING: CPU: 2 PID: 11122 at mm/page_alloc.c:2989 __alloc_pages_nodemask+0x695/0x14e0() Call Trace: [< inline >] __dump_stack lib/dump_stack.c:15 [<ffffffff82743b56>] dump_stack+0x68/0x92 lib/dump_stack.c:50 [<ffffffff81244ec9>] warn_slowpath_common+0xd9/0x140 kernel/panic.c:460 [<ffffffff812450f9>] warn_slowpath_null+0x29/0x30 kernel/panic.c:493 [< inline >] __alloc_pages_slowpath mm/page_alloc.c:2989 [<ffffffff81554e95>] __alloc_pages_nodemask+0x695/0x14e0 mm/page_alloc.c:3235 [<ffffffff816188fe>] alloc_pages_current+0xee/0x340 mm/mempolicy.c:2055 [< inline >] alloc_pages include/linux/gfp.h:451 [<ffffffff81550706>] alloc_kmem_pages+0x16/0xf0 mm/page_alloc.c:3414 [<ffffffff815a1c89>] kmalloc_order+0x19/0x60 mm/slab_common.c:1007 [<ffffffff815a1cef>] kmalloc_order_trace+0x1f/0xa0 mm/slab_common.c:1018 [< inline >] kmalloc_large include/linux/slab.h:390 [<ffffffff81627784>] __kmalloc+0x234/0x250 mm/slub.c:3525 [< inline >] kmalloc include/linux/slab.h:463 [< inline >] map_update_elem kernel/bpf/syscall.c:288 [< inline >] SYSC_bpf kernel/bpf/syscall.c:744 To avoid never succeeding kmalloc with order >= MAX_ORDER check that elem->value_size and computed elem_size are within limits for both hash and array type maps. Also add __GFP_NOWARN to kmalloc(value_size | elem_size) to avoid OOM warnings. Note kmalloc(key_size) is highly unlikely to trigger OOM, since key_size <= 512, so keep those kmalloc-s as-is. Large value_size can cause integer overflows in elem_size and map.pages formulas, so check for that as well. Fixes: aaac3ba95e4c ("bpf: charge user for creation of BPF maps and programs") Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-11-30 00:59:35 +00:00
MAX_BPF_STACK - sizeof(struct htab_elem))
/* if value_size is bigger, the user space won't be able to
* access the elements via bpf syscall. This check also makes
* sure that the elem_size doesn't overflow and it's
* kmalloc-able later in htab_map_update_elem()
*/
goto free_htab;
if (percpu && round_up(htab->map.value_size, 8) > PCPU_MIN_UNIT_SIZE)
/* make sure the size for pcpu_alloc() is reasonable */
goto free_htab;
bpf: fix allocation warnings in bpf maps and integer overflow For large map->value_size the user space can trigger memory allocation warnings like: WARNING: CPU: 2 PID: 11122 at mm/page_alloc.c:2989 __alloc_pages_nodemask+0x695/0x14e0() Call Trace: [< inline >] __dump_stack lib/dump_stack.c:15 [<ffffffff82743b56>] dump_stack+0x68/0x92 lib/dump_stack.c:50 [<ffffffff81244ec9>] warn_slowpath_common+0xd9/0x140 kernel/panic.c:460 [<ffffffff812450f9>] warn_slowpath_null+0x29/0x30 kernel/panic.c:493 [< inline >] __alloc_pages_slowpath mm/page_alloc.c:2989 [<ffffffff81554e95>] __alloc_pages_nodemask+0x695/0x14e0 mm/page_alloc.c:3235 [<ffffffff816188fe>] alloc_pages_current+0xee/0x340 mm/mempolicy.c:2055 [< inline >] alloc_pages include/linux/gfp.h:451 [<ffffffff81550706>] alloc_kmem_pages+0x16/0xf0 mm/page_alloc.c:3414 [<ffffffff815a1c89>] kmalloc_order+0x19/0x60 mm/slab_common.c:1007 [<ffffffff815a1cef>] kmalloc_order_trace+0x1f/0xa0 mm/slab_common.c:1018 [< inline >] kmalloc_large include/linux/slab.h:390 [<ffffffff81627784>] __kmalloc+0x234/0x250 mm/slub.c:3525 [< inline >] kmalloc include/linux/slab.h:463 [< inline >] map_update_elem kernel/bpf/syscall.c:288 [< inline >] SYSC_bpf kernel/bpf/syscall.c:744 To avoid never succeeding kmalloc with order >= MAX_ORDER check that elem->value_size and computed elem_size are within limits for both hash and array type maps. Also add __GFP_NOWARN to kmalloc(value_size | elem_size) to avoid OOM warnings. Note kmalloc(key_size) is highly unlikely to trigger OOM, since key_size <= 512, so keep those kmalloc-s as-is. Large value_size can cause integer overflows in elem_size and map.pages formulas, so check for that as well. Fixes: aaac3ba95e4c ("bpf: charge user for creation of BPF maps and programs") Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-11-30 00:59:35 +00:00
htab->elem_size = sizeof(struct htab_elem) +
round_up(htab->map.key_size, 8);
if (percpu)
htab->elem_size += sizeof(void *);
else
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
htab->elem_size += round_up(htab->map.value_size, 8);
bpf: fix allocation warnings in bpf maps and integer overflow For large map->value_size the user space can trigger memory allocation warnings like: WARNING: CPU: 2 PID: 11122 at mm/page_alloc.c:2989 __alloc_pages_nodemask+0x695/0x14e0() Call Trace: [< inline >] __dump_stack lib/dump_stack.c:15 [<ffffffff82743b56>] dump_stack+0x68/0x92 lib/dump_stack.c:50 [<ffffffff81244ec9>] warn_slowpath_common+0xd9/0x140 kernel/panic.c:460 [<ffffffff812450f9>] warn_slowpath_null+0x29/0x30 kernel/panic.c:493 [< inline >] __alloc_pages_slowpath mm/page_alloc.c:2989 [<ffffffff81554e95>] __alloc_pages_nodemask+0x695/0x14e0 mm/page_alloc.c:3235 [<ffffffff816188fe>] alloc_pages_current+0xee/0x340 mm/mempolicy.c:2055 [< inline >] alloc_pages include/linux/gfp.h:451 [<ffffffff81550706>] alloc_kmem_pages+0x16/0xf0 mm/page_alloc.c:3414 [<ffffffff815a1c89>] kmalloc_order+0x19/0x60 mm/slab_common.c:1007 [<ffffffff815a1cef>] kmalloc_order_trace+0x1f/0xa0 mm/slab_common.c:1018 [< inline >] kmalloc_large include/linux/slab.h:390 [<ffffffff81627784>] __kmalloc+0x234/0x250 mm/slub.c:3525 [< inline >] kmalloc include/linux/slab.h:463 [< inline >] map_update_elem kernel/bpf/syscall.c:288 [< inline >] SYSC_bpf kernel/bpf/syscall.c:744 To avoid never succeeding kmalloc with order >= MAX_ORDER check that elem->value_size and computed elem_size are within limits for both hash and array type maps. Also add __GFP_NOWARN to kmalloc(value_size | elem_size) to avoid OOM warnings. Note kmalloc(key_size) is highly unlikely to trigger OOM, since key_size <= 512, so keep those kmalloc-s as-is. Large value_size can cause integer overflows in elem_size and map.pages formulas, so check for that as well. Fixes: aaac3ba95e4c ("bpf: charge user for creation of BPF maps and programs") Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-11-30 00:59:35 +00:00
/* prevent zero size kmalloc and check for u32 overflow */
if (htab->n_buckets == 0 ||
htab->n_buckets > U32_MAX / sizeof(struct bucket))
goto free_htab;
cost = (u64) htab->n_buckets * sizeof(struct bucket) +
(u64) htab->elem_size * htab->map.max_entries;
if (percpu)
cost += (u64) round_up(htab->map.value_size, 8) *
num_possible_cpus() * htab->map.max_entries;
else
cost += (u64) htab->elem_size * num_possible_cpus();
if (cost >= U32_MAX - PAGE_SIZE)
bpf: fix allocation warnings in bpf maps and integer overflow For large map->value_size the user space can trigger memory allocation warnings like: WARNING: CPU: 2 PID: 11122 at mm/page_alloc.c:2989 __alloc_pages_nodemask+0x695/0x14e0() Call Trace: [< inline >] __dump_stack lib/dump_stack.c:15 [<ffffffff82743b56>] dump_stack+0x68/0x92 lib/dump_stack.c:50 [<ffffffff81244ec9>] warn_slowpath_common+0xd9/0x140 kernel/panic.c:460 [<ffffffff812450f9>] warn_slowpath_null+0x29/0x30 kernel/panic.c:493 [< inline >] __alloc_pages_slowpath mm/page_alloc.c:2989 [<ffffffff81554e95>] __alloc_pages_nodemask+0x695/0x14e0 mm/page_alloc.c:3235 [<ffffffff816188fe>] alloc_pages_current+0xee/0x340 mm/mempolicy.c:2055 [< inline >] alloc_pages include/linux/gfp.h:451 [<ffffffff81550706>] alloc_kmem_pages+0x16/0xf0 mm/page_alloc.c:3414 [<ffffffff815a1c89>] kmalloc_order+0x19/0x60 mm/slab_common.c:1007 [<ffffffff815a1cef>] kmalloc_order_trace+0x1f/0xa0 mm/slab_common.c:1018 [< inline >] kmalloc_large include/linux/slab.h:390 [<ffffffff81627784>] __kmalloc+0x234/0x250 mm/slub.c:3525 [< inline >] kmalloc include/linux/slab.h:463 [< inline >] map_update_elem kernel/bpf/syscall.c:288 [< inline >] SYSC_bpf kernel/bpf/syscall.c:744 To avoid never succeeding kmalloc with order >= MAX_ORDER check that elem->value_size and computed elem_size are within limits for both hash and array type maps. Also add __GFP_NOWARN to kmalloc(value_size | elem_size) to avoid OOM warnings. Note kmalloc(key_size) is highly unlikely to trigger OOM, since key_size <= 512, so keep those kmalloc-s as-is. Large value_size can cause integer overflows in elem_size and map.pages formulas, so check for that as well. Fixes: aaac3ba95e4c ("bpf: charge user for creation of BPF maps and programs") Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-11-30 00:59:35 +00:00
/* make sure page count doesn't overflow */
goto free_htab;
htab->map.pages = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
bpf: fix allocation warnings in bpf maps and integer overflow For large map->value_size the user space can trigger memory allocation warnings like: WARNING: CPU: 2 PID: 11122 at mm/page_alloc.c:2989 __alloc_pages_nodemask+0x695/0x14e0() Call Trace: [< inline >] __dump_stack lib/dump_stack.c:15 [<ffffffff82743b56>] dump_stack+0x68/0x92 lib/dump_stack.c:50 [<ffffffff81244ec9>] warn_slowpath_common+0xd9/0x140 kernel/panic.c:460 [<ffffffff812450f9>] warn_slowpath_null+0x29/0x30 kernel/panic.c:493 [< inline >] __alloc_pages_slowpath mm/page_alloc.c:2989 [<ffffffff81554e95>] __alloc_pages_nodemask+0x695/0x14e0 mm/page_alloc.c:3235 [<ffffffff816188fe>] alloc_pages_current+0xee/0x340 mm/mempolicy.c:2055 [< inline >] alloc_pages include/linux/gfp.h:451 [<ffffffff81550706>] alloc_kmem_pages+0x16/0xf0 mm/page_alloc.c:3414 [<ffffffff815a1c89>] kmalloc_order+0x19/0x60 mm/slab_common.c:1007 [<ffffffff815a1cef>] kmalloc_order_trace+0x1f/0xa0 mm/slab_common.c:1018 [< inline >] kmalloc_large include/linux/slab.h:390 [<ffffffff81627784>] __kmalloc+0x234/0x250 mm/slub.c:3525 [< inline >] kmalloc include/linux/slab.h:463 [< inline >] map_update_elem kernel/bpf/syscall.c:288 [< inline >] SYSC_bpf kernel/bpf/syscall.c:744 To avoid never succeeding kmalloc with order >= MAX_ORDER check that elem->value_size and computed elem_size are within limits for both hash and array type maps. Also add __GFP_NOWARN to kmalloc(value_size | elem_size) to avoid OOM warnings. Note kmalloc(key_size) is highly unlikely to trigger OOM, since key_size <= 512, so keep those kmalloc-s as-is. Large value_size can cause integer overflows in elem_size and map.pages formulas, so check for that as well. Fixes: aaac3ba95e4c ("bpf: charge user for creation of BPF maps and programs") Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-11-30 00:59:35 +00:00
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
/* if map size is larger than memlock limit, reject it early */
err = bpf_map_precharge_memlock(htab->map.pages);
if (err)
goto free_htab;
bpf: fix allocation warnings in bpf maps and integer overflow For large map->value_size the user space can trigger memory allocation warnings like: WARNING: CPU: 2 PID: 11122 at mm/page_alloc.c:2989 __alloc_pages_nodemask+0x695/0x14e0() Call Trace: [< inline >] __dump_stack lib/dump_stack.c:15 [<ffffffff82743b56>] dump_stack+0x68/0x92 lib/dump_stack.c:50 [<ffffffff81244ec9>] warn_slowpath_common+0xd9/0x140 kernel/panic.c:460 [<ffffffff812450f9>] warn_slowpath_null+0x29/0x30 kernel/panic.c:493 [< inline >] __alloc_pages_slowpath mm/page_alloc.c:2989 [<ffffffff81554e95>] __alloc_pages_nodemask+0x695/0x14e0 mm/page_alloc.c:3235 [<ffffffff816188fe>] alloc_pages_current+0xee/0x340 mm/mempolicy.c:2055 [< inline >] alloc_pages include/linux/gfp.h:451 [<ffffffff81550706>] alloc_kmem_pages+0x16/0xf0 mm/page_alloc.c:3414 [<ffffffff815a1c89>] kmalloc_order+0x19/0x60 mm/slab_common.c:1007 [<ffffffff815a1cef>] kmalloc_order_trace+0x1f/0xa0 mm/slab_common.c:1018 [< inline >] kmalloc_large include/linux/slab.h:390 [<ffffffff81627784>] __kmalloc+0x234/0x250 mm/slub.c:3525 [< inline >] kmalloc include/linux/slab.h:463 [< inline >] map_update_elem kernel/bpf/syscall.c:288 [< inline >] SYSC_bpf kernel/bpf/syscall.c:744 To avoid never succeeding kmalloc with order >= MAX_ORDER check that elem->value_size and computed elem_size are within limits for both hash and array type maps. Also add __GFP_NOWARN to kmalloc(value_size | elem_size) to avoid OOM warnings. Note kmalloc(key_size) is highly unlikely to trigger OOM, since key_size <= 512, so keep those kmalloc-s as-is. Large value_size can cause integer overflows in elem_size and map.pages formulas, so check for that as well. Fixes: aaac3ba95e4c ("bpf: charge user for creation of BPF maps and programs") Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-11-30 00:59:35 +00:00
err = -ENOMEM;
htab->buckets = bpf_map_area_alloc(htab->n_buckets *
sizeof(struct bucket));
if (!htab->buckets)
goto free_htab;
for (i = 0; i < htab->n_buckets; i++) {
INIT_HLIST_NULLS_HEAD(&htab->buckets[i].head, i);
raw_spin_lock_init(&htab->buckets[i].lock);
}
if (prealloc) {
err = prealloc_init(htab);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
if (err)
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
goto free_buckets;
if (!percpu && !lru) {
/* lru itself can remove the least used element, so
* there is no need for an extra elem during map_update.
*/
err = alloc_extra_elems(htab);
if (err)
goto free_prealloc;
}
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
}
return &htab->map;
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
free_prealloc:
prealloc_destroy(htab);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
free_buckets:
bpf_map_area_free(htab->buckets);
free_htab:
kfree(htab);
return ERR_PTR(err);
}
static inline u32 htab_map_hash(const void *key, u32 key_len)
{
return jhash(key, key_len, 0);
}
static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
{
return &htab->buckets[hash & (htab->n_buckets - 1)];
}
static inline struct hlist_nulls_head *select_bucket(struct bpf_htab *htab, u32 hash)
{
return &__select_bucket(htab, hash)->head;
}
/* this lookup function can only be called with bucket lock taken */
static struct htab_elem *lookup_elem_raw(struct hlist_nulls_head *head, u32 hash,
void *key, u32 key_size)
{
struct hlist_nulls_node *n;
struct htab_elem *l;
hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
if (l->hash == hash && !memcmp(&l->key, key, key_size))
return l;
return NULL;
}
/* can be called without bucket lock. it will repeat the loop in
* the unlikely event when elements moved from one bucket into another
* while link list is being walked
*/
static struct htab_elem *lookup_nulls_elem_raw(struct hlist_nulls_head *head,
u32 hash, void *key,
u32 key_size, u32 n_buckets)
{
struct hlist_nulls_node *n;
struct htab_elem *l;
again:
hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
if (l->hash == hash && !memcmp(&l->key, key, key_size))
return l;
if (unlikely(get_nulls_value(n) != (hash & (n_buckets - 1))))
goto again;
return NULL;
}
/* Called from syscall or from eBPF program directly, so
* arguments have to match bpf_map_lookup_elem() exactly.
* The return value is adjusted by BPF instructions
* in htab_map_gen_lookup().
*/
static void *__htab_map_lookup_elem(struct bpf_map *map, void *key)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct hlist_nulls_head *head;
struct htab_elem *l;
u32 hash, key_size;
/* Must be called with rcu_read_lock. */
WARN_ON_ONCE(!rcu_read_lock_held());
key_size = map->key_size;
hash = htab_map_hash(key, key_size);
head = select_bucket(htab, hash);
l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
return l;
}
static void *htab_map_lookup_elem(struct bpf_map *map, void *key)
{
struct htab_elem *l = __htab_map_lookup_elem(map, key);
if (l)
return l->key + round_up(map->key_size, 8);
return NULL;
}
/* inline bpf_map_lookup_elem() call.
* Instead of:
* bpf_prog
* bpf_map_lookup_elem
* map->ops->map_lookup_elem
* htab_map_lookup_elem
* __htab_map_lookup_elem
* do:
* bpf_prog
* __htab_map_lookup_elem
*/
static u32 htab_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
{
struct bpf_insn *insn = insn_buf;
const int ret = BPF_REG_0;
*insn++ = BPF_EMIT_CALL((u64 (*)(u64, u64, u64, u64, u64))__htab_map_lookup_elem);
*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
*insn++ = BPF_ALU64_IMM(BPF_ADD, ret,
offsetof(struct htab_elem, key) +
round_up(map->key_size, 8));
return insn - insn_buf;
}
static void *htab_lru_map_lookup_elem(struct bpf_map *map, void *key)
{
struct htab_elem *l = __htab_map_lookup_elem(map, key);
if (l) {
bpf_lru_node_set_ref(&l->lru_node);
return l->key + round_up(map->key_size, 8);
}
return NULL;
}
/* It is called from the bpf_lru_list when the LRU needs to delete
* older elements from the htab.
*/
static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node)
{
struct bpf_htab *htab = (struct bpf_htab *)arg;
struct htab_elem *l = NULL, *tgt_l;
struct hlist_nulls_head *head;
struct hlist_nulls_node *n;
unsigned long flags;
struct bucket *b;
tgt_l = container_of(node, struct htab_elem, lru_node);
b = __select_bucket(htab, tgt_l->hash);
head = &b->head;
raw_spin_lock_irqsave(&b->lock, flags);
hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
if (l == tgt_l) {
hlist_nulls_del_rcu(&l->hash_node);
break;
}
raw_spin_unlock_irqrestore(&b->lock, flags);
return l == tgt_l;
}
/* Called from syscall */
static int htab_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct hlist_nulls_head *head;
struct htab_elem *l, *next_l;
u32 hash, key_size;
int i = 0;
WARN_ON_ONCE(!rcu_read_lock_held());
key_size = map->key_size;
if (!key)
goto find_first_elem;
hash = htab_map_hash(key, key_size);
head = select_bucket(htab, hash);
/* lookup the key */
l = lookup_nulls_elem_raw(head, hash, key, key_size, htab->n_buckets);
if (!l)
goto find_first_elem;
/* key was found, get next key in the same bucket */
next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_next_rcu(&l->hash_node)),
struct htab_elem, hash_node);
if (next_l) {
/* if next elem in this hash list is non-zero, just return it */
memcpy(next_key, next_l->key, key_size);
return 0;
}
/* no more elements in this hash list, go to the next bucket */
i = hash & (htab->n_buckets - 1);
i++;
find_first_elem:
/* iterate over buckets */
for (; i < htab->n_buckets; i++) {
head = select_bucket(htab, i);
/* pick first element in the bucket */
next_l = hlist_nulls_entry_safe(rcu_dereference_raw(hlist_nulls_first_rcu(head)),
struct htab_elem, hash_node);
if (next_l) {
/* if it's not empty, just return it */
memcpy(next_key, next_l->key, key_size);
return 0;
}
}
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
/* iterated over all buckets and all elements */
return -ENOENT;
}
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
static void htab_elem_free(struct bpf_htab *htab, struct htab_elem *l)
{
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
if (htab->map.map_type == BPF_MAP_TYPE_PERCPU_HASH)
free_percpu(htab_elem_get_ptr(l, htab->map.key_size));
kfree(l);
}
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
static void htab_elem_free_rcu(struct rcu_head *head)
{
struct htab_elem *l = container_of(head, struct htab_elem, rcu);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
struct bpf_htab *htab = l->htab;
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
/* must increment bpf_prog_active to avoid kprobe+bpf triggering while
* we're calling kfree, otherwise deadlock is possible if kprobes
* are placed somewhere inside of slub
*/
preempt_disable();
__this_cpu_inc(bpf_prog_active);
htab_elem_free(htab, l);
__this_cpu_dec(bpf_prog_active);
preempt_enable();
}
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)
{
struct bpf_map *map = &htab->map;
if (map->ops->map_fd_put_ptr) {
void *ptr = fd_htab_map_get_ptr(map, l);
map->ops->map_fd_put_ptr(ptr);
}
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
if (htab_is_prealloc(htab)) {
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
pcpu_freelist_push(&htab->freelist, &l->fnode);
} else {
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
atomic_dec(&htab->count);
l->htab = htab;
call_rcu(&l->rcu, htab_elem_free_rcu);
}
}
static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
void *value, bool onallcpus)
{
if (!onallcpus) {
/* copy true value_size bytes */
memcpy(this_cpu_ptr(pptr), value, htab->map.value_size);
} else {
u32 size = round_up(htab->map.value_size, 8);
int off = 0, cpu;
for_each_possible_cpu(cpu) {
bpf_long_memcpy(per_cpu_ptr(pptr, cpu),
value + off, size);
off += size;
}
}
}
static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
void *value, u32 key_size, u32 hash,
bool percpu, bool onallcpus,
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
struct htab_elem *old_elem)
{
u32 size = htab->map.value_size;
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
bool prealloc = htab_is_prealloc(htab);
struct htab_elem *l_new, **pl_new;
void __percpu *pptr;
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
if (prealloc) {
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
if (old_elem) {
/* if we're updating the existing element,
* use per-cpu extra elems to avoid freelist_pop/push
*/
pl_new = this_cpu_ptr(htab->extra_elems);
l_new = *pl_new;
*pl_new = old_elem;
} else {
struct pcpu_freelist_node *l;
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
l = pcpu_freelist_pop(&htab->freelist);
if (!l)
return ERR_PTR(-E2BIG);
l_new = container_of(l, struct htab_elem, fnode);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
}
} else {
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
if (atomic_inc_return(&htab->count) > htab->map.max_entries)
if (!old_elem) {
/* when map is full and update() is replacing
* old element, it's ok to allocate, since
* old element will be freed immediately.
* Otherwise return an error
*/
atomic_dec(&htab->count);
return ERR_PTR(-E2BIG);
}
l_new = kmalloc(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN);
if (!l_new)
return ERR_PTR(-ENOMEM);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
}
memcpy(l_new->key, key, key_size);
if (percpu) {
/* round up value_size to 8 bytes */
size = round_up(size, 8);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
if (prealloc) {
pptr = htab_elem_get_ptr(l_new, key_size);
} else {
/* alloc_percpu zero-fills */
pptr = __alloc_percpu_gfp(size, 8,
GFP_ATOMIC | __GFP_NOWARN);
if (!pptr) {
kfree(l_new);
return ERR_PTR(-ENOMEM);
}
}
pcpu_copy_value(htab, pptr, value, onallcpus);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
if (!prealloc)
htab_elem_set_ptr(l_new, key_size, pptr);
} else {
memcpy(l_new->key + round_up(key_size, 8), value, size);
}
l_new->hash = hash;
return l_new;
}
static int check_flags(struct bpf_htab *htab, struct htab_elem *l_old,
u64 map_flags)
{
if (l_old && map_flags == BPF_NOEXIST)
/* elem already exists */
return -EEXIST;
if (!l_old && map_flags == BPF_EXIST)
/* elem doesn't exist, cannot update it */
return -ENOENT;
return 0;
}
/* Called from syscall or from eBPF program */
static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
u64 map_flags)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct htab_elem *l_new = NULL, *l_old;
struct hlist_nulls_head *head;
unsigned long flags;
struct bucket *b;
u32 key_size, hash;
int ret;
if (unlikely(map_flags > BPF_EXIST))
/* unknown flags */
return -EINVAL;
WARN_ON_ONCE(!rcu_read_lock_held());
key_size = map->key_size;
hash = htab_map_hash(key, key_size);
b = __select_bucket(htab, hash);
head = &b->head;
/* bpf_map_update_elem() can be called in_irq() */
raw_spin_lock_irqsave(&b->lock, flags);
l_old = lookup_elem_raw(head, hash, key, key_size);
ret = check_flags(htab, l_old, map_flags);
if (ret)
goto err;
l_new = alloc_htab_elem(htab, key, value, key_size, hash, false, false,
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
l_old);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
if (IS_ERR(l_new)) {
/* all pre-allocated elements are in use or memory exhausted */
ret = PTR_ERR(l_new);
goto err;
}
/* add new element to the head of the list, so that
* concurrent search will find it before old elem
*/
hlist_nulls_add_head_rcu(&l_new->hash_node, head);
if (l_old) {
hlist_nulls_del_rcu(&l_old->hash_node);
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
if (!htab_is_prealloc(htab))
free_htab_elem(htab, l_old);
}
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
ret = 0;
err:
raw_spin_unlock_irqrestore(&b->lock, flags);
return ret;
}
static int htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value,
u64 map_flags)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct htab_elem *l_new, *l_old = NULL;
struct hlist_nulls_head *head;
unsigned long flags;
struct bucket *b;
u32 key_size, hash;
int ret;
if (unlikely(map_flags > BPF_EXIST))
/* unknown flags */
return -EINVAL;
WARN_ON_ONCE(!rcu_read_lock_held());
key_size = map->key_size;
hash = htab_map_hash(key, key_size);
b = __select_bucket(htab, hash);
head = &b->head;
/* For LRU, we need to alloc before taking bucket's
* spinlock because getting free nodes from LRU may need
* to remove older elements from htab and this removal
* operation will need a bucket lock.
*/
l_new = prealloc_lru_pop(htab, key, hash);
if (!l_new)
return -ENOMEM;
memcpy(l_new->key + round_up(map->key_size, 8), value, map->value_size);
/* bpf_map_update_elem() can be called in_irq() */
raw_spin_lock_irqsave(&b->lock, flags);
l_old = lookup_elem_raw(head, hash, key, key_size);
ret = check_flags(htab, l_old, map_flags);
if (ret)
goto err;
/* add new element to the head of the list, so that
* concurrent search will find it before old elem
*/
hlist_nulls_add_head_rcu(&l_new->hash_node, head);
if (l_old) {
bpf_lru_node_set_ref(&l_new->lru_node);
hlist_nulls_del_rcu(&l_old->hash_node);
}
ret = 0;
err:
raw_spin_unlock_irqrestore(&b->lock, flags);
if (ret)
bpf_lru_push_free(&htab->lru, &l_new->lru_node);
else if (l_old)
bpf_lru_push_free(&htab->lru, &l_old->lru_node);
return ret;
}
static int __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
void *value, u64 map_flags,
bool onallcpus)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct htab_elem *l_new = NULL, *l_old;
struct hlist_nulls_head *head;
unsigned long flags;
struct bucket *b;
u32 key_size, hash;
int ret;
if (unlikely(map_flags > BPF_EXIST))
/* unknown flags */
return -EINVAL;
WARN_ON_ONCE(!rcu_read_lock_held());
key_size = map->key_size;
hash = htab_map_hash(key, key_size);
b = __select_bucket(htab, hash);
head = &b->head;
/* bpf_map_update_elem() can be called in_irq() */
raw_spin_lock_irqsave(&b->lock, flags);
l_old = lookup_elem_raw(head, hash, key, key_size);
ret = check_flags(htab, l_old, map_flags);
if (ret)
goto err;
if (l_old) {
/* per-cpu hash map can update value in-place */
pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
value, onallcpus);
} else {
l_new = alloc_htab_elem(htab, key, value, key_size,
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
hash, true, onallcpus, NULL);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
if (IS_ERR(l_new)) {
ret = PTR_ERR(l_new);
goto err;
}
hlist_nulls_add_head_rcu(&l_new->hash_node, head);
}
ret = 0;
err:
raw_spin_unlock_irqrestore(&b->lock, flags);
return ret;
}
static int __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
void *value, u64 map_flags,
bool onallcpus)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct htab_elem *l_new = NULL, *l_old;
struct hlist_nulls_head *head;
unsigned long flags;
struct bucket *b;
u32 key_size, hash;
int ret;
if (unlikely(map_flags > BPF_EXIST))
/* unknown flags */
return -EINVAL;
WARN_ON_ONCE(!rcu_read_lock_held());
key_size = map->key_size;
hash = htab_map_hash(key, key_size);
b = __select_bucket(htab, hash);
head = &b->head;
/* For LRU, we need to alloc before taking bucket's
* spinlock because LRU's elem alloc may need
* to remove older elem from htab and this removal
* operation will need a bucket lock.
*/
if (map_flags != BPF_EXIST) {
l_new = prealloc_lru_pop(htab, key, hash);
if (!l_new)
return -ENOMEM;
}
/* bpf_map_update_elem() can be called in_irq() */
raw_spin_lock_irqsave(&b->lock, flags);
l_old = lookup_elem_raw(head, hash, key, key_size);
ret = check_flags(htab, l_old, map_flags);
if (ret)
goto err;
if (l_old) {
bpf_lru_node_set_ref(&l_old->lru_node);
/* per-cpu hash map can update value in-place */
pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
value, onallcpus);
} else {
pcpu_copy_value(htab, htab_elem_get_ptr(l_new, key_size),
value, onallcpus);
hlist_nulls_add_head_rcu(&l_new->hash_node, head);
l_new = NULL;
}
ret = 0;
err:
raw_spin_unlock_irqrestore(&b->lock, flags);
if (l_new)
bpf_lru_push_free(&htab->lru, &l_new->lru_node);
return ret;
}
static int htab_percpu_map_update_elem(struct bpf_map *map, void *key,
void *value, u64 map_flags)
{
return __htab_percpu_map_update_elem(map, key, value, map_flags, false);
}
static int htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
void *value, u64 map_flags)
{
return __htab_lru_percpu_map_update_elem(map, key, value, map_flags,
false);
}
/* Called from syscall or from eBPF program */
static int htab_map_delete_elem(struct bpf_map *map, void *key)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct hlist_nulls_head *head;
struct bucket *b;
struct htab_elem *l;
unsigned long flags;
u32 hash, key_size;
int ret = -ENOENT;
WARN_ON_ONCE(!rcu_read_lock_held());
key_size = map->key_size;
hash = htab_map_hash(key, key_size);
b = __select_bucket(htab, hash);
head = &b->head;
raw_spin_lock_irqsave(&b->lock, flags);
l = lookup_elem_raw(head, hash, key, key_size);
if (l) {
hlist_nulls_del_rcu(&l->hash_node);
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
free_htab_elem(htab, l);
ret = 0;
}
raw_spin_unlock_irqrestore(&b->lock, flags);
return ret;
}
static int htab_lru_map_delete_elem(struct bpf_map *map, void *key)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct hlist_nulls_head *head;
struct bucket *b;
struct htab_elem *l;
unsigned long flags;
u32 hash, key_size;
int ret = -ENOENT;
WARN_ON_ONCE(!rcu_read_lock_held());
key_size = map->key_size;
hash = htab_map_hash(key, key_size);
b = __select_bucket(htab, hash);
head = &b->head;
raw_spin_lock_irqsave(&b->lock, flags);
l = lookup_elem_raw(head, hash, key, key_size);
if (l) {
hlist_nulls_del_rcu(&l->hash_node);
ret = 0;
}
raw_spin_unlock_irqrestore(&b->lock, flags);
if (l)
bpf_lru_push_free(&htab->lru, &l->lru_node);
return ret;
}
static void delete_all_elements(struct bpf_htab *htab)
{
int i;
for (i = 0; i < htab->n_buckets; i++) {
struct hlist_nulls_head *head = select_bucket(htab, i);
struct hlist_nulls_node *n;
struct htab_elem *l;
hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
hlist_nulls_del_rcu(&l->hash_node);
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
htab_elem_free(htab, l);
}
}
}
/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
static void htab_map_free(struct bpf_map *map)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
/* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
* so the programs (can be more than one that used this map) were
* disconnected from events. Wait for outstanding critical sections in
* these programs to complete
*/
synchronize_rcu();
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
/* some of free_htab_elem() callbacks for elements of this map may
* not have executed. Wait for them.
*/
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
rcu_barrier();
bpf: fix hashmap extra_elems logic In both kmalloc and prealloc mode the bpf_map_update_elem() is using per-cpu extra_elems to do atomic update when the map is full. There are two issues with it. The logic can be misused, since it allows max_entries+num_cpus elements to be present in the map. And alloc_extra_elems() at map creation time can fail percpu alloc for large map values with a warn: WARNING: CPU: 3 PID: 2752 at ../mm/percpu.c:892 pcpu_alloc+0x119/0xa60 illegal size (32824) or align (8) for percpu allocation The fixes for both of these issues are different for kmalloc and prealloc modes. For prealloc mode allocate extra num_possible_cpus elements and store their pointers into extra_elems array instead of actual elements. Hence we can use these hidden(spare) elements not only when the map is full but during bpf_map_update_elem() that replaces existing element too. That also improves performance, since pcpu_freelist_pop/push is avoided. Unfortunately this approach cannot be used for kmalloc mode which needs to kfree elements after rcu grace period. Therefore switch it back to normal kmalloc even when full and old element exists like it was prior to commit 6c9059817432 ("bpf: pre-allocate hash map elements"). Add tests to check for over max_entries and large map values. Reported-by: Dave Jones <davej@codemonkey.org.uk> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements") Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 02:05:04 +00:00
if (!htab_is_prealloc(htab))
bpf: pre-allocate hash map elements If kprobe is placed on spin_unlock then calling kmalloc/kfree from bpf programs is not safe, since the following dead lock is possible: kfree->spin_lock(kmem_cache_node->lock)...spin_unlock->kprobe-> bpf_prog->map_update->kmalloc->spin_lock(of the same kmem_cache_node->lock) and deadlocks. The following solutions were considered and some implemented, but eventually discarded - kmem_cache_create for every map - add recursion check to slow-path of slub - use reserved memory in bpf_map_update for in_irq or in preempt_disabled - kmalloc via irq_work At the end pre-allocation of all map elements turned out to be the simplest solution and since the user is charged upfront for all the memory, such pre-allocation doesn't affect the user space visible behavior. Since it's impossible to tell whether kprobe is triggered in a safe location from kmalloc point of view, use pre-allocation by default and introduce new BPF_F_NO_PREALLOC flag. While testing of per-cpu hash maps it was discovered that alloc_percpu(GFP_ATOMIC) has odd corner cases and often fails to allocate memory even when 90% of it is free. The pre-allocation of per-cpu hash elements solves this problem as well. Turned out that bpf_map_update() quickly followed by bpf_map_lookup()+bpf_map_delete() is very common pattern used in many of iovisor/bcc/tools, so there is additional benefit of pre-allocation, since such use cases are must faster. Since all hash map elements are now pre-allocated we can remove atomic increment of htab->count and save few more cycles. Also add bpf_map_precharge_memlock() to check rlimit_memlock early to avoid large malloc/free done by users who don't have sufficient limits. Pre-allocation is done with vmalloc and alloc/free is done via percpu_freelist. Here are performance numbers for different pre-allocation algorithms that were implemented, but discarded in favor of percpu_freelist: 1 cpu: pcpu_ida 2.1M pcpu_ida nolock 2.3M bt 2.4M kmalloc 1.8M hlist+spinlock 2.3M pcpu_freelist 2.6M 4 cpu: pcpu_ida 1.5M pcpu_ida nolock 1.8M bt w/smp_align 1.7M bt no/smp_align 1.1M kmalloc 0.7M hlist+spinlock 0.2M pcpu_freelist 2.0M 8 cpu: pcpu_ida 0.7M bt w/smp_align 0.8M kmalloc 0.4M pcpu_freelist 1.5M 32 cpu: kmalloc 0.13M pcpu_freelist 0.49M pcpu_ida nolock is a modified percpu_ida algorithm without percpu_ida_cpu locks and without cross-cpu tag stealing. It's faster than existing percpu_ida, but not as fast as pcpu_freelist. bt is a variant of block/blk-mq-tag.c simlified and customized for bpf use case. bt w/smp_align is using cache line for every 'long' (similar to blk-mq-tag). bt no/smp_align allocates 'long' bitmasks continuously to save memory. It's comparable to percpu_ida and in some cases faster, but slower than percpu_freelist hlist+spinlock is the simplest free list with single spinlock. As expeceted it has very bad scaling in SMP. kmalloc is existing implementation which is still available via BPF_F_NO_PREALLOC flag. It's significantly slower in single cpu and in 8 cpu setup it's 3 times slower than pre-allocation with pcpu_freelist, but saves memory, so in cases where map->max_entries can be large and number of map update/delete per second is low, it may make sense to use it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-03-08 05:57:15 +00:00
delete_all_elements(htab);
else
prealloc_destroy(htab);
free_percpu(htab->extra_elems);
bpf_map_area_free(htab->buckets);
kfree(htab);
}
const struct bpf_map_ops htab_map_ops = {
.map_alloc = htab_map_alloc,
.map_free = htab_map_free,
.map_get_next_key = htab_map_get_next_key,
.map_lookup_elem = htab_map_lookup_elem,
.map_update_elem = htab_map_update_elem,
.map_delete_elem = htab_map_delete_elem,
.map_gen_lookup = htab_map_gen_lookup,
};
const struct bpf_map_ops htab_lru_map_ops = {
.map_alloc = htab_map_alloc,
.map_free = htab_map_free,
.map_get_next_key = htab_map_get_next_key,
.map_lookup_elem = htab_lru_map_lookup_elem,
.map_update_elem = htab_lru_map_update_elem,
.map_delete_elem = htab_lru_map_delete_elem,
};
/* Called from eBPF program */
static void *htab_percpu_map_lookup_elem(struct bpf_map *map, void *key)
{
struct htab_elem *l = __htab_map_lookup_elem(map, key);
if (l)
return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
else
return NULL;
}
static void *htab_lru_percpu_map_lookup_elem(struct bpf_map *map, void *key)
{
struct htab_elem *l = __htab_map_lookup_elem(map, key);
if (l) {
bpf_lru_node_set_ref(&l->lru_node);
return this_cpu_ptr(htab_elem_get_ptr(l, map->key_size));
}
return NULL;
}
int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct htab_elem *l;
void __percpu *pptr;
int ret = -ENOENT;
int cpu, off = 0;
u32 size;
/* per_cpu areas are zero-filled and bpf programs can only
* access 'value_size' of them, so copying rounded areas
* will not leak any kernel data
*/
size = round_up(map->value_size, 8);
rcu_read_lock();
l = __htab_map_lookup_elem(map, key);
if (!l)
goto out;
if (htab_is_lru(htab))
bpf_lru_node_set_ref(&l->lru_node);
pptr = htab_elem_get_ptr(l, map->key_size);
for_each_possible_cpu(cpu) {
bpf_long_memcpy(value + off,
per_cpu_ptr(pptr, cpu), size);
off += size;
}
ret = 0;
out:
rcu_read_unlock();
return ret;
}
int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
u64 map_flags)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
int ret;
rcu_read_lock();
if (htab_is_lru(htab))
ret = __htab_lru_percpu_map_update_elem(map, key, value,
map_flags, true);
else
ret = __htab_percpu_map_update_elem(map, key, value, map_flags,
true);
rcu_read_unlock();
return ret;
}
const struct bpf_map_ops htab_percpu_map_ops = {
.map_alloc = htab_map_alloc,
.map_free = htab_map_free,
.map_get_next_key = htab_map_get_next_key,
.map_lookup_elem = htab_percpu_map_lookup_elem,
.map_update_elem = htab_percpu_map_update_elem,
.map_delete_elem = htab_map_delete_elem,
};
const struct bpf_map_ops htab_lru_percpu_map_ops = {
.map_alloc = htab_map_alloc,
.map_free = htab_map_free,
.map_get_next_key = htab_map_get_next_key,
.map_lookup_elem = htab_lru_percpu_map_lookup_elem,
.map_update_elem = htab_lru_percpu_map_update_elem,
.map_delete_elem = htab_lru_map_delete_elem,
};
static struct bpf_map *fd_htab_map_alloc(union bpf_attr *attr)
{
struct bpf_map *map;
if (attr->value_size != sizeof(u32))
return ERR_PTR(-EINVAL);
/* pointer is stored internally */
attr->value_size = sizeof(void *);
map = htab_map_alloc(attr);
attr->value_size = sizeof(u32);
return map;
}
static void fd_htab_map_free(struct bpf_map *map)
{
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
struct hlist_nulls_node *n;
struct hlist_nulls_head *head;
struct htab_elem *l;
int i;
for (i = 0; i < htab->n_buckets; i++) {
head = select_bucket(htab, i);
hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
void *ptr = fd_htab_map_get_ptr(map, l);
map->ops->map_fd_put_ptr(ptr);
}
}
htab_map_free(map);
}
/* only called from syscall */
int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
void *key, void *value, u64 map_flags)
{
void *ptr;
int ret;
u32 ufd = *(u32 *)value;
ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
if (IS_ERR(ptr))
return PTR_ERR(ptr);
ret = htab_map_update_elem(map, key, &ptr, map_flags);
if (ret)
map->ops->map_fd_put_ptr(ptr);
return ret;
}
static struct bpf_map *htab_of_map_alloc(union bpf_attr *attr)
{
struct bpf_map *map, *inner_map_meta;
inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
if (IS_ERR(inner_map_meta))
return inner_map_meta;
map = fd_htab_map_alloc(attr);
if (IS_ERR(map)) {
bpf_map_meta_free(inner_map_meta);
return map;
}
map->inner_map_meta = inner_map_meta;
return map;
}
static void *htab_of_map_lookup_elem(struct bpf_map *map, void *key)
{
struct bpf_map **inner_map = htab_map_lookup_elem(map, key);
if (!inner_map)
return NULL;
return READ_ONCE(*inner_map);
}
static void htab_of_map_free(struct bpf_map *map)
{
bpf_map_meta_free(map->inner_map_meta);
fd_htab_map_free(map);
}
const struct bpf_map_ops htab_of_maps_map_ops = {
.map_alloc = htab_of_map_alloc,
.map_free = htab_of_map_free,
.map_get_next_key = htab_map_get_next_key,
.map_lookup_elem = htab_of_map_lookup_elem,
.map_delete_elem = htab_map_delete_elem,
.map_fd_get_ptr = bpf_map_fd_get_ptr,
.map_fd_put_ptr = bpf_map_fd_put_ptr,
};